Research On Short-term Load Forecasting Of Power System Based On Virtual Forecasting

At present,China’s economic development is in a new normal state.Electric power reforms are advancing vertically.There are large fluctuations and uncertainties in power load growth.Load forecasting can provide reliable decision-making basis for grid planning and is important for efficient grid operation.Short-term load forecasting is an important part of load forecasting,and is of great significance for improving the utilization rate of power generation equipment and economic dispatch.Based on the evaluation and comparison of traditional load forecasting methods,this paper introduces the idea of virtual forecasting and applies it to the research of short-term load forecasting of power systems.First of all,with the increasing trend of power load big data as the background,fully considering the historical load and various meteorological factors,in order to solve the phenomenon of high-dimensional data information mixing,the relevant data is filtered and mined:wavelet threshold denoising technology is used to process historical load data,Repair abnormal data;Introduce generalized grey absolute correlation as the screening index of meteorological factors,select 6 factors from 12 meteorological factors,and then use PCA algorithm to perform dimension reduction processing and feature extraction on the 6 meteorological factors to obtain effective meteorological data Sample set.Secondly,using BP neural network,support vector machine,gray prediction and other different prediction methods to make virtual predictions for the same historical period,and using the maximum dispersion comprehensive attribute value as the evaluation index,four of the six single-item models were selected and selected.Species model.Finally,with the aim of minimizing the sum of squared virtual prediction residuals,a comprehensive model of short-term load forecasting under the optimal virtual prediction mode is established,and the model is used to predict the true "To-Be-Predicted Day" to verify the rationality of the model and algorithm And effectiveness.